Sysfs is too restrictive

Sysfs is too restrictive

[Stephen Hemminger]

I have been experimenting with using sysfs to manage Ethernet bridge devices
but have to conclude that sysfs is too restrictive to be useful for doing
it.  Maybe for 2.7, some of the implementation restrictions that tie kobjects 
directly to the file hierarchy can be lifted.

Here is what I want to do.  Assume there are only two kobject's related to
bridging.  The first is the class device (inside the net_device) that contains
"br0".  The other is the bridge port which represents an interface in a bridge.
The hack method of adding kobjects to create subdirectories would make the already
painful ref count issues worse.

It is too hard to get the desired directory layout without re-implementing most
of sysfs in the driver.  This is what I want, and don't have Pat to kick around
anymore.

class
`-- net
    |-- br0
    |   |-- addr_len
    |   |-- address
    |   |-- bridge
    |   |   |-- forward_delay
    |   |   |-- hello_time
    |   |   |-- id
    |   |   |-- max_age
    |   |   |-- port
    |   |   |   |-- cost
    |   |   |   |-- eth0 -> ../../../eth0
    |   |   |   |-- priority
    |   |   |   `-- stp
    |   |   `-- priority
    |   |-- broadcast
    |   |-- features
    |   |-- flags
    |   |-- ifindex
    |   |-- mtu
    |   |-- statistics
    |   |   |-- collisions
    |   |   |-- multicast
    |   |   |-- rx_bytes
    |   |   |-- rx_compressed
    |   |   |-- rx_crc_errors
    |   |   |-- rx_dropped
    |   |   |-- rx_errors
    |   |   |-- rx_fifo_errors
    |   |   |-- rx_frame_errors
    |   |   |-- rx_length_errors
    |   |   |-- rx_missed_errors
    |   |   |-- rx_over_errors
    |   |   |-- rx_packets
    |   |   |-- tx_aborted_errors
    |   |   |-- tx_bytes
    |   |   |-- tx_carrier_errors
    |   |   |-- tx_compressed
    |   |   |-- tx_dropped
    |   |   |-- tx_errors
    |   |   |-- tx_fifo_errors
    |   |   |-- tx_heartbeat_errors
    |   |   |-- tx_packets
    |   |   `-- tx_window_errors
    |   |-- tx_queue_len
    |   `-- type
    `-- eth0
        |-- addr_len
        |-- address
        |-- broadcast
        |-- device -> ../../../devices/pci0000:00/0000:00:1e.0/0000:04:04.0
        |-- driver -> ../../../bus/pci/drivers/e100
        |-- features
        |-- flags
        |-- ifindex
        |-- iflink
        |-- mtu
        |-- statistics
        |   |-- collisions
        |   |-- multicast
        |   |-- rx_bytes
        |   |-- rx_compressed
        |   |-- rx_crc_errors
        |   |-- rx_dropped
        |   |-- rx_errors
        |   |-- rx_fifo_errors
        |   |-- rx_frame_errors
        |   |-- rx_length_errors
        |   |-- rx_missed_errors
        |   |-- rx_over_errors
        |   |-- rx_packets
        |   |-- tx_aborted_errors
        |   |-- tx_bytes
        |   |-- tx_carrier_errors
        |   |-- tx_compressed
        |   |-- tx_dropped
        |   |-- tx_errors
        |   |-- tx_fifo_errors
        |   |-- tx_heartbeat_errors
        |   |-- tx_packets
        |   `-- tx_window_errors
        |-- tx_queue_len
        `-- type

8 directories, 75 files

Re: Sysfs is too restrictive

[Dave Jones]

On Fri, Feb 27, 2004 at 10:05:41AM -0800, Stephen Hemminger wrote:
 >     |   |-- bridge
 >     |   |   |-- forward_delay
 >     |   |   |-- hello_time
 >     |   |   |-- id
 >     |   |   |-- max_age
 >     |   |   |-- port
 >     |   |   |   |-- cost
 >     |   |   |   |-- eth0 -> ../../../eth0
 >     |   |   |   |-- priority
 >     |   |   |   `-- stp
 >     |   |   `-- priority
 >     |   |-- broadcast

Shouldn't you be seeing the other side of the bridge in here too ?
Ie, if br0 is a bridge between eth0 and eth1, how does that fit
your plan ?

		Dave

Re: Sysfs is too restrictive

[Stephen Hemminger]

On Fri, 27 Feb 2004 18:14:55 +0000
Dave Jones <davej@redhat.com> wrote:

> On Fri, Feb 27, 2004 at 10:05:41AM -0800, Stephen Hemminger wrote:
>  >     |   |-- bridge
>  >     |   |   |-- forward_delay
>  >     |   |   |-- hello_time
>  >     |   |   |-- id
>  >     |   |   |-- max_age
>  >     |   |   |-- port
>  >     |   |   |   |-- cost
>  >     |   |   |   |-- eth0 -> ../../../eth0
>  >     |   |   |   |-- priority
>  >     |   |   |   `-- stp
>  >     |   |   `-- priority
>  >     |   |-- broadcast
> 
> Shouldn't you be seeing the other side of the bridge in here too ?
> Ie, if br0 is a bridge between eth0 and eth1, how does that fit
> your plan ?

yes, there would be multiple interfaces in the port directory.
More like this:

br0
`-- bridge
    |-- forward_delay
    |-- id
    |-- max_age
    |-- port
    |   |-- eth0
    |   |   |-- cost
    |   |   |-- interface -> ../../../../eth0
    |   |   |-- priority
    |   |   `-- stp
    |   `-- eth1
    |   |   |-- cost
    |   |   |-- interface -> ../../../../eth1
    |   |   |-- priority
    |   |   `-- stp
    `-- priority

4 directories, 8 files

Re: Sysfs is too restrictive

[Dave Jones]

On Fri, Feb 27, 2004 at 10:30:03AM -0800, Stephen Hemminger wrote:

 > > Shouldn't you be seeing the other side of the bridge in here too ?
 > > Ie, if br0 is a bridge between eth0 and eth1, how does that fit
 > > your plan ?
 > 
 > yes, there would be multiple interfaces in the port directory.
 > More like this:

Ok, that makes more sense.

		Dave

Re: Sysfs is too restrictive

[Greg KH]

On Fri, Feb 27, 2004 at 10:05:41AM -0800, Stephen Hemminger wrote:
> 
> It is too hard to get the desired directory layout without re-implementing most
> of sysfs in the driver.  This is what I want, and don't have Pat to kick around
> anymore.

Yes, I agree that it needs to be made much easier to create a
subdirectory in sysfs using the driver model.  I even filed a bug about
that a long time ago :)

It's on my TODO list for 2.7, if that makes you feel any better.  And if
you want to try to implement it before I get to it, that would be great.

Sorry I don't have a solution for you to do this with today :(

greg k-h